Crab (Callinectes sapidus) Meat by Electronic Nose and Draeger-Tube Analysis

Page created by Angel Lewis
 
CONTINUE READING
Determination of Quality Attributes of Blue
1547-0636
1049-8850
WAFP
Journal of Aquatic Food Product Technology
                                Technology, Vol. 17, No. 3, June 2008: pp. 1–29

       Crab (Callinectes sapidus) Meat by
   Electronic Nose and Draeger-Tube Analysis
                                                                                         Paul J. Sarnoski
Sarnoski et OF
JOURNAL     al. AQUATIC FOOD PRODUCT TECHNOLOGY

                                                                                        Michael L. Jahncke
                                                                                         Sean F. O’Keefe
                                                                                  Parameswarakumar Mallikarjunan
                                                                                        George J. Flick, Jr.

                            ABSTRACT. In this study, five groups of sequentially spoiled crabmeat
                            were evaluated by a trained sensory panel, and these results were compared
                            with the findings from a Cyranose 320 Electronic Nose and Draeger gas

   Paul J. Sarnoski, Sean F. O’Keefe, and George J. Flick, Jr. are affiliated with
the Department of Food Science and Technology, Virginia Polytechnic
Institute and State University, Blacksburg, VA, USA.
    Michael L. Jahncke is affiliated with the Department of Food Science and
Technology and the Virginia Seafood Agricultural Research and Extension
Center, Hampton, VA, USA.
    Parameswarakumar Mallikarjunan is affiliated with the Department of
Biological Systems Engineering, Virginia Polytechnic Institute and State University,
Blacksburg, VA, USA.
    Luis F. Martinez and Dr. Murat Balaban from the University of Florida are
recognized for recommending the compressed air method. The idea of using com-
pressed breathing air that cycles into sample containers was first developed at the
University of Florida. A similar setup was used for this study. Funding for this
research was provided by the Virginia Sea Grant Program, the World Food Logis-
tics Organization, the International Association of Refrigerated Warehouses
(WFLO/IARW), and the National Fisheries Institute (NFI) Scholarship Program.
    Address correspondence to: Michael L. Jahncke, Department of Food Science and
Technology and the Virginia Seafood Agricultural Research and Extension Center,
102 South King Street, Hampton, VA 23669 (E-mail: mjahncke@vt.edu).
                                            Journal of Aquatic Food Product Technology, Vol. 17(3) 2008
                                                  Available online at http://jafpt.haworthpress.com
                                                 © 2008 by The Haworth Press. All rights reserved.
234                                                      doi:10.1080/10498850802183364
Sarnoski et al.                               235

     analyzer. Using the electronic nose with filtered compressed breathing air
     yielded the best results. Although this approach resulted in 100 % separa-
     tion of the known groups, only 30% of the coded unknown samples were
     correctly identified. All 5 groups of samples analyzed using Draeger-Tubes
     were found to be significantly different at α=0.05 using a Tukey-Kramer
     ANOVA statistical procedure. The coded unknown samples were correctly
     identified at a rate of 83%. The simplicity and precision of this latter proce-
     dure may present opportunities for use of Draeger-Tubes by crab process-
     ing industries and other food processing industries as an objective method
     for quality control.

     KEYWORDS. Electronic nose, blue crab, Draeger-tube, quality, ammonia,
     sensory

                             INTRODUCTION

   The blue crab industry represents a major resource for the American
seafood industry. The blue crab (Callinectes sapidus) is fished commer-
cially from Maryland to south of Galveston, Texas; however the highest
population of the species resides in the Chesapeake Bay and its tributar-
ies. From 1968 to 2004 the annual harvest was 73 million pounds in the
Chesapeake Bay alone (Bonzek et al., 2005). In recent years the harvests
have dipped into the 50 million pound range. However, in 2004, the
harvest was approximately 60 million pounds. The 2004 harvest trans-
lated to approximately 55 million dollars in revenue. The industry is
important to the economy of the Chesapeake Bay region, and care needs
to be taken to ensure that over-crabbing, habitat destruction, and other
adverse environmental conditions do not lead to extinction of the blue
crab.
   The spoilage of crustaceans involves metabolic activities of microor-
ganisms and enzymes. All of these processes are interrelated and occur
simultaneously or at different stages of spoilage. The mechanism for the
chemical breakdown of crabmeat is not as well established as for some
other types of seafood, such as fish. However, it is believed the chemical
breakdown is relatively similar to that of other shellfish, but not fish, as
crabmeat has a higher free amino acid composition compared to fish.
High levels of taurine, proline, glycine, alanine, and arginine are found in
crustaceans (Martin et al., 1982). Other nitrogenous compounds, such as
peptides and other nonprotein nitrogen components found in shellfish,
236       JOURNAL OF AQUATIC FOOD PRODUCT TECHNOLOGY

form typical spoilage metabolites such as indole, ammonia, putrescine,
histamine, and cadaverine (Eskin et al., 1971). An increase in tissue
ammonia levels during spoilage is attributed to several enzymatic pro-
cesses: deamination of free amino acids, degradation of nucleotides, and
oxidation of amines (Gill, 1990). Since ammonia and other volatile
amines have been implicated in seafood spoilage, it is believed that being
able to measure their levels relatively quickly and easily would be of great
advantage in determining quality levels.
   Amines are typically produced in shellfish by the bacterial deamination
or decarboxylation of free amino acids. Deamination of amino acids
results in the formation of ammonia and organic acids, while decarboxy-
lation leads to the production of amines and carbon dioxide. Breakdown
of trimethylamine oxide (TMAO) may also result in the formation of tri-
methylamine (TMA) and dimethylamine (DMA) in certain seafood prod-
ucts (Martin et al., 1982). These compounds are part of a total volatile
base (TVB) analysis, and methods have been derived to determine TMA
and DMA concentrations separately (Jones et al., 1998; Yerlikaya and
Gökoðlu, 2004). In an analysis of cod fish by Gill (1990), the relationship
between TMA and subjective evaluation of odor was approximately
linear.
   Finding an economical way to evaluate seafood quality is difficult.
Quality can be evaluated using trained certified sensory experts, but these
experts are expensive to train, and periodic recertification is needed if fed-
eral and state agencies are to accept them as experts. Sensory evaluation
also is expensive and difficult for the industry due to frequent personnel
turnover and the extensive time commitments needed to maintain certi-
fied sensory personnel.
   A possible solution to this problem is to develop a method to determine
quality using electronic instrumentation. Possible instrumentation that
could be used to determine quality includes the electronic nose. The elec-
tronic nose has been used in a variety of applications, including food
applications (Arshak et al., 2004), and is a rapid technique with analysis
time often taking less than 2 minutes.
   The electronic nose uses an array of sensors designed to mimic the
mammalian olfactory system’s response to aromas. Instead of receptor
proteins, which are part of the human olfactory system, the electronic
nose in most cases uses a sensor array system. The odor molecules are
drawn into the electronic nose using sampling techniques such as head-
space sampling, diffusion methods, bubblers, or preconcentrators (Pearce
et al., 2003). The sample odors are drawn across the sensor array, which
Sarnoski et al.                         237

induces a reversible physical or chemical change in the sensing material,
in turn causing a change in electrical properties (Harsanyi, 2000). Each
sensor in an array behaves in a similar manner to a mammalian protein
receptor responding to different odors to varying degrees. The sensor sig-
nal is transduced into electrical signals, which are preprocessed and con-
ditioned (usually a normalization technique is used) before being
identified by a pattern recognition system.
   Other methods for determining spoilage in seafood products include ana-
lyzing for ammonia and volatile amines. An AOAC-approved method for
the colorimetric determination of ammonia in crabmeat exists (Steinbrecher,
1973), but this procedure is laborious and time-consuming. Ammonia can
also be measured by diffusion, paper test strips, and ammonia selective
electrodes, but these methods are often less sensitive compared with the
colorimetric method and have interference issues, such as sodium inter-
fering with an ammonia-sensitive electrode. The established methods
require an aqueous extract. Volatile amines are also a good indicator of
spoilage in seafood as total volatile base (TVB) analysis is often used to
determine spoilage. In that regard, ammonia-sensitive Draeger-Tubes®,
used mostly for the detection of ammonia gas leaks, may work as a rapid
analytical test to determine spoilage in crabmeat. The tubes have cross-
sensitivities with other basic substances, such as organic amines that are
likewise indicated but have differing sensitivities (Dräger-Rohrchen,
2001). Since a no-specific test has previously proved somewhat effective
(TVB analysis), the cross sensitivity to other amines may be advanta-
geous for determining spoilage in shellfish.

                   MATERIALS AND METHODS

Blue Crab Samples for Electronic Nose and Draeger-Tube
Analysis
  Experiments were conducted with body meat (i.e., regular or special)
obtained from a local crab processing company (Hampton, VA, USA) and
harvested from the Chesapeake Bay during the same harvest season.
Eight separate batches of crabmeat were spoiled (136–181 kgs/batch) and
used to train the sensory panel. One of these spoilage batches was also
used to train the electronic nose and one was used to establish the stan-
dard curve for the Draeger-Tube analyses. To validate the electronic nose
and Draeger-Tubes, an additional batch of crabmeat was sequentially
238       JOURNAL OF AQUATIC FOOD PRODUCT TECHNOLOGY

spoiled and used to verify the accuracy and precision of the analytical
methods. The batches were separated into individual spoilage increments
(18 kg/increment) and then vacuum packed in Kapak (Minneapolis, MN,
USA) SealPac pouches, 0.47 liter (2.5 mil PET/LLDPE). Reference (0
day) samples were first frozen at –18°C for 12–20 hours and then stored
at –80°C. The remaining samples were sequentially spoiled at ambient
(21°C) and refrigerated storage conditions (4°C), frozen at –18°C for 12–20
hours, and then stored at –80°C until needed for analysis. The following
samples were analyzed by the electronic nose: reference (0 day); 6 and 9 h
ambient; and 7, 8, and 9 day cold samples. Draeger-Tube (Dräger-Röhrchen,
Luebeck, Germany) analyses were conducted on reference samples
(0 day) and samples held for 4, 5, 6, and 7 days at 4°C. Samples were
thawed overnight in a refrigerator (4°C) before analysis.

Microbial and Sensory Analysis
   Fresh-picked crabmeat was collected from the processing plant and
held on ice at refrigeration temperatures (< 4.4°C). The initial bacterial
load was enumerated by mesophilic aerobic plate count using standard
AOAC-approved methods as listed in the FDA Bacteriological Analytical
Manual (FDA, 2001). Twenty-five gram portions of well-mixed crabmeat
were blended with 225 ml sterile 0.1 % Peptone buffer in a filter stoma-
cher bag and stomached for 2 min at 230 RPM for a 1:10 dilution. Subse-
quent dilutions were prepared by adding 1 ml to 9 ml of sterile Peptone
broth. One ml portions were plated on 3M Petrifilm aerobic count plates
and incubated at 35°C for 48 hours. Aerobic plate counts (APC) were
reported as colony forming units per gram (CFU/g).
   Crabmeat was divided into 5 lb. increments and spread into 10 by 16 in.
lidded plastic containers for controlled decomposition at 21°C (ambient
temperature) or 4°C (refrigerated storage). Crabmeat was periodically
examined for sensory quality, packaged in 100–200 gram portions in
Kapak 402 bags (1 pt. size, 6 1/2 in. by 8 in.), vacuum sealed, labeled, and
frozen overnight at –18°C before placement at -80°C for long-term stor-
age. Fifty grams were used for bacteriological examination, as described
above, for each spoilage increment. Reference samples were immediately
bagged, frozen, and stored.
   Spoiled crabmeat was evaluated for odor, taste, and texture against
reference samples using a 6-member trained sensory panel. The sensory
panelists underwent training for 8–12 months using the Spectrum™
Descriptive Analysis Method (Meilgaard et al., 1999) to evaluate the
Sarnoski et al.                         239

attributes of coded unknown blue crabmeat samples spoiled under refrig-
erated (4°C) and ambient conditions (21°C). A 10 cm line scale was used
to evaluate the sensory attributes of the crabmeat samples with a score
higher than 5.0 cm classified as a fail. Samples with scores higher than
2.5 cm but less than 5.0 cm were classified as borderline pass (mid-pass).
Similarly, samples with scores lower than 7.5 cm but higher than 5.0 cm
were classified as borderline fail (mid-fail). Samples were removed from
–80°C storage and thawed overnight in a refrigerator (4°C). Packages
were opened and the crabmeat was placed into 2 oz. lidded cups. The cups
were heated for 20 seconds at 50% power in a 1000-watt microwave oven
and immediately served to panelists. Reference samples representing dif-
ferent food intensity attributes were provided to the panelists prior to
evaluating the crabmeat samples. Sensory and microbiological analyses
were conducted at the Virginia Seafood Agricultural Research and Exten-
sion Center, Hampton, VA, USA.
Electronic Nose Analysis
   Thawed 25-gram samples were placed into individual sealed 200-ml
glass jars (Ball Corporation, Broomfield, CO, USA), which then were
placed in a water bath at 40°C for 30 minutes. All samples reached equi-
librium temperatures to allow headspace odors to accumulate.
Manufacturer’s Recommended Method
   Electronic nose analysis used individually sealed glass jars for head-
space analysis. A hole was punched into the lid top to allow an entry point
for the insertion of the electronic nose sniffing needle. A small piece of
sticky foam (Darice Inc., Strongsville, OH, USA) was purchased from the
local craft store and used as a septum. The electronic nose used for this
analysis was a Cyranose320™ (Smiths Detection, Pasadena, CA, USA),
equipped with 32 polymer-conducting sensors. Ten samples of known
quality crabmeat were used to build the training set for each class, and 20
separate coded crabmeat samples were analyzed to validate the system.
The operating conditions used were described in the manual. The settings
used were as follows: the baseline purge 10 seconds, sample draw one 25
seconds, first air intake purge 10 seconds, and second sample gas purge
75 seconds. Digital filtering and normalization were used; the substrate
heater was set to 45°C. Pump speed was set to high (180 cc/min). After
analysis of each sample, a 1-min equilibration period allowed the sensors
to equilibrate before being exposed to another sample.
240       JOURNAL OF AQUATIC FOOD PRODUCT TECHNOLOGY

Compressed Air Method
    Compressed tank breathing air (Airgas Mid America, Radford, VA,
USA) filtered through activated carbon and moisture traps cycled through
a reference and sample container was used. One of the outlets on the ref-
erence container was connected to the purge inlet of the electronic nose to
ensure the nose was receiving clean, dry air that was consistent on a day-
to-day basis. The flow rate through this apparatus was 15 ml/sec. A tube
connected the sample inlet of the electronic nose to the sample container.
Ten samples were used to build the training set for each class, and 20
coded crabmeat samples were used to validate the system. Samples were
not reanalyzed. The inside of the sample container was cleaned between
samples. The settings used were as follows: the baseline purge 15 seconds,
sample draw one 30 seconds, first sample gas purge 10 seconds, first air
intake purge 10 seconds, and second sample gas purge 30 seconds. Digital
filtering and normalization were used; the substrate heater was set to
45°C. Pump speed was set to medium (120 cc/min). After each sample
analysis, a 2-min equilibration period allowed the sensors to equilibrate
before being exposed to another sample.
Crab Reference Method
   The same settings used for the compressed air method were used for
the crab reference method. The reference (0 day) crabmeat sample was
placed in the reference container connected to the purge inlet. The flow
rate through the apparatus was 15 ml/sec. Ten samples were used to build
the training set for each class, and 20 coded crabmeat samples were used
to validate the system. Samples were not reanalyzed. Both the sample and
reference containers were cleaned between samples. A 2-min equilibra-
tion period allowed the sensors to equilibrate before being exposed to
another sample.

Draeger-Tube Analysis
   Samples were thawed in the refrigerator (4°C) overnight. Samples
were brought to 40°C in a bag using a water bath. Bags were stomached
for 1 min at 230 RPM. The 25-gram samples were then transferred to a
sealed glass jar. Headspace was allowed to accumulate for 10 min. Sam-
ples were then analyzed using 2–30 ppm, or 5–70 ppm ammonia-sensitive
Draeger-Tubes. The Accuro® (Dräger-Röhrchen, Luebeck, Germany) gas
detector pump was used to pull 100 ml of sample through the short-term
Sarnoski et al.                          241

detector tubes. Ten known samples were used to construct a standard
curve using 2 replications for each spoilage group. Then validation of the
trial was performed using unknown samples.
Statistical Analysis
   Electronic nose data was processed using the PC Nose software
version 6.5 (Smiths Detection, Pasadena, CA, USA), JMP IN 5.1 (SAS
Institute, Cary, NC, USA), and SAS 9.1 (SAS Institute, Cary, NC, USA).
The built-in PC Nose software contains a program that automatically runs
canonical and principal component analysis (PCA). In addition, the raw
sensor response data from the electronic nose was processed using statis-
tical software (JMP and SAS). Canonical discriminant analysis (CDA)
and stepwise discriminant analysis (SDA) were performed using SAS
products. For all three methods, the following sensors were omitted from
the model for stepwise discriminant analysis: 4–6, 11, 14–16, 19, 21, 22,
24, 26, 28, 30, and 31. Stepwise variable selection was done to eliminate
sensors that showed correlation with other sensors (data) in the model.
Analysis of variance (ANOVA) for the Draeger-Tube results was
performed using JMP.

                   RESULTS AND DISCUSSION

Electronic Nose Analysis
   The electronic nose has been used for food quality applications
(Chantarachoti et al., 2006; Hu et al., 2005; O’Connell et al., 2001).
Van Deventer and Mallikarjunan (2002) found the Cyranose 320 con-
ducting polymer electronic nose to be superior to metal oxide and quartz
crystal microbalance systems, which is why this particular electronic nose
was selected for this study. For the compressed air and crab reference
methods, filtered, clean, dry air from the same tank and a filtering proce-
dure were used. The manufacturer recommends using ambient air for the
test; however, it is possible that results can be affected by the varying
overall quality and composition of ambient air on a daily basis. It was
hypothesized that using filtered, clean, dry air rather than ambient labora-
tory air may provide more accurate results by eliminating sensor drift,
thus achieving a more stable baseline. In addition to being able to classify
differences between groups through electronic nose training procedures,
validation of the model is needed in order to demonstrate the accuracy
242       JOURNAL OF AQUATIC FOOD PRODUCT TECHNOLOGY

and precision of the electronic nose and Draeger-Tubes. This was accom-
plished by using known samples from multiple spoilage trials to train the
electronic nose and to establish the standard curve for the Draeger-Tube
analyses. The final 2 spoilage trials used coded samples to validate the
electronic nose and Draeger-Tube analyses, respectively.
   The compressed air method achieved 100% correct separation of all 5
groups of sequentially spoiled crabmeat as verified by a cross-validation
technique (see Figure 1). This technique treats each individual sample as an
unknown and then re-enters it into the statistical model. A cross-validation
statistic of 100% shows each sample belongs to only one group. A poor
cross validation statistic means that one or more samples can be included
in multiple groups, thus decreasing the ability to distinguish between
groups. For the compressed air method, all groups were significantly dif-
ferent at α = 0.05 using CDA and SDA. In fact, all groups were signifi-
cantly different (p < 0.0001). Separation of different known samples was
excellent using the compressed air method, but identification of coded

FIGURE 1. PC nose PCA projection plot for training set data for the
compressed air method.
Sarnoski et al.                           243

samples was not as accurate. Only 30% of the coded samples were cor-
rectly identified using CDA and SDA models. The PCA model was even
less effective at correctly identifying coded samples; only 20% of the
coded samples were correctly classified using this model (see Table 1).
   The manufacturer’s recommended method correctly separated 30% of
known sample groups using PCA analysis (see Figure 2). Better separa-
tion was achieved using CDA analysis (Table 1). Correct group separa-
tion was improved to 44% using CDA analysis, and 50% using SDA
analysis, respectively. Classification of coded samples of crabmeat was
low using the manufacturer’s recommended method. SDA analysis
yielded the best results, correctly identifying 25% of coded samples. PCA
and CDA analysis correctly identified only 10% and 20%, respectively, of
the coded crabmeat samples.
   The crab reference method used reference crabmeat (high quality
meat) as a baseline for comparison of spoiled samples to unspoiled sam-
ples. Only 50% of the samples were correctly separated using PCA as the
statistical method. CDA was able to correctly separate 68% of the groups,
while SDA was able to correctly separate 92% of the groups at α = 0.05.
CDA eliminates sensors that show correlation with other sensors from the
model; in this case using this statistical method seemed to improve group
separation. Correlation between electronic nose sensors was most likely
high due to the samples analyzed having similar attributes. Sensory scores
indicated that degrees of spoilage between samples were similar; however
there is no tangible way to link the electronic nose sensor response to sen-
sory data because the relationship of specific chemical odorants to spe-
cific sensors is proprietary. However, the crab reference method should

        TABLE 1. Comparative results for electronic nose studies

Setup                  Statistical                Group             Correct
                        Method                 Separation (%)   Identification (%)

Manufacturer’s           PCA                            30             10
 Recommended             CDA                            44             20
                         SDA                            50             25
Compressed Air           PCA                           100             20
                         CDA                           100             30
                         SDA                           100             30
Crab Reference           PCA                            50             20
                         CDA                            68             35
                         SDA                            92             20
244        JOURNAL OF AQUATIC FOOD PRODUCT TECHNOLOGY

FIGURE 2. PC nose PCA projection plot for training set data for the
manufacturer’s recommended method.

have accentuated differences due to spoilage, since spoilage attributes
should have been the only differences between the samples. For the coded
crabmeat samples, CDA classified 35% of samples correctly. The PCA
and SDA results were less accurate, correctly identifying only 20% of the
coded samples. It was expected that SDA analysis on the coded crabmeat
samples would result in the highest correct classification method, since
for the previous analyses, separation and identification of unknown sam-
ples were highest using this statistical method.
   The compressed air method had the highest success rate in separating and
identifying the coded samples. The manufacturer’s recommended analysis
procedure produced the poorest results for separation and identification of the
coded crabmeat samples. These results suggest that methods using clean dry
tank air will produce better group separation. However, low rates of correct
classification of coded samples were also obtained using clean, dry, filtered air.
   ANOVA analysis and declassification of some of the groups were also
conducted. ANOVA analysis of individual sensors in the 32-sensor array
was conducted to see if one sensor could be used to determine spoilage
Sarnoski et al.                            245

levels. In these analyses, one sensor was not more effective than PCA,
CDA, or SDA for correctly identifying spoilage levels. Another technique
employed was to group together the 6 and 9 h ambient samples, the 7 and
9 day cold samples, and reference samples (0 day) forming a 3-group
model. Sensory evaluations indicate that the ambient increment and cold
spoiled increment had similar attributes regardless of spoilage time (see
Table 2). However, combining the groups did not considerably improve
the accuracy of the unknown identification results. Water may have inter-
fered with the sensors, confounding the results. When the water-sensitive
sensors, numbers 5, 6, 23, and 31 according to The Practical Guide to the
Cyranose 320™ (Cyrano Sciences, 2001), were omitted from the model,
group separation was still poor. Perhaps these sensors are also sensitive to
the polar components present in the aroma profile of spoiling crabmeat.
   In order to gain insight into why the coded sample results differed from
the training data using known samples, the coded sample results were
plotted using the training data of the compressed air method (see Figure
3). Most of the coded samples were identified as low quality even though
high quality (0 day) samples were present in the coded sample set. None
of the 0 day samples were correctly identified. There may have been a
“hold over” affect from the more spoiled samples that affected the results
of the less spoiled samples, even though the sample containers were
cleaned after each trial and the sensors were purged according to the pre-
scribed settings of the Cyranose 320 manual. Since samples were of similar
quality (according to microbial and sensory analysis) for both known and
unknown samples, differences between the training data and the coded
sample data may be due to deficiencies in the sensor purging system of
the electronic nose.
   The electronic nose may not have been sufficiently sensitive to distin-
guish between minor differences between classes of spoiled crabmeat.

           TABLE 2. Aerobic plate counts and sensory results
           for samples used for the Cyranose 320™ analysis

          Sample              APC (CFU/g)        Sensory Classification

          Reference (0 day)     9.2 • 104        High pass
          6 hr. ambient         4.6 • 105        Mid-pass
          9 hr. ambient         1.0 • 106        Mid-pass/low pass
          7 day cold            7.5 • 106        Low fail/mid-fail
          9 day cold            4.0 • 107        Mid-fail/high fail
246                    JOURNAL OF AQUATIC FOOD PRODUCT TECHNOLOGY

FIGURE 3. Compressed air method CDA plot with coded unknown
samples plotted.

                                                                       9hramb
                                                         6hramb
               80

               75
                        ref

               70
  Canonical2

               65
                                     7daycold

                                                                                     9daycold
               60

               55

               50
                 –15      –10   –5              0   5      10     15            20              25   30
                                                    Canonical1

The crabmeat samples were of varying degrees of spoilage, which to the
electronic nose may not have been perceived as a noticeable difference.
Ammonia, which is associated with crabmeat spoilage, is soluble in
water. This may have led to ammonia not being at a detectable level by
the electronic nose, since crabmeat also contains an appreciable amount
of water. In addition, no recommended calibration procedure exists for
this electronic nose. It is also possible that as the conducting polymer sen-
sors age, sensor drift may be an issue with electronic nose devices (Hines
et al., 1999). These issues have been shown to occur when the electronic
nose has been used to quantitatively identify components (Harper, 2001).
This study is similar to a quantification study, since as spoilage increases,
the intensity of spoilage attributes (odors) increase.
   Suggestions for improving the electronic nose results include using
fewer sample groups, bootstrapping, and neural networks. Instead of
using 5 groups for training, use a high quality group and a low quality
group. Bootstrapping is an approach best utilized for validating a given
model, especially when a small sample size is used. Bootstrapping creates
a larger sample size by using real data to create imaginary data. For exam-
ple, 30 real observations can create 1000 “bootstrap” observations. This
approach is best when the error in analysis is attributed to small sample
Sarnoski et al.                         247

size. The bootstrap method was used in a study of beef spoilage, and it
improved the classification accuracy from 87% to 98% (Balasubramanian
et al., 2004).
   Artificial neural networks are mathematical creations inspired by the
biological nervous system. An artificial neural network consists of a
lattice of information processing elements called neurons, which are con-
nected together in a certain way (known as the architecture). The
strengths of these connections are called synaptic weights. These weights
are determined either during a training (or learning) phase for supervised
neural networks or by an algorithm for unsupervised neural networks
(Gardner and Bartlett, 1999). Therefore, a neural network can “learn” as
data is imputed in, whereas a pattern recognition technique simply looks
for patterns. The use of neural networks significantly improved pattern
recognition for electronic nose systems (Zhang et al., 2003) and was
found to be effective for quantifying different spice mixture compositions
(Zhang et al., 2005). Due to time constraints, this technique was not
applied to this data set.

Draeger-Tube Analysis
   The data from the sequentially spoiled samples was plotted in a standard
curve, using an average of 2 replications for each group (see Figure 4).
These data show an expected pattern: as the product spoils, the ppm value
for ammonia increases. These data fit an exponential equation rather
than a linear equation. This is not surprising since microbial growth fol-
lows an exponential equation (Marr, 2000). ANOVA analysis was con-
ducted on the Draeger-Tube results (see Figure 5). Means were separated
using a Tukey-Kramer statistical test.
   Coded samples were analyzed using the Draeger-Tube method
(See Table 3). Correct classification of 83% of the coded samples was
achieved. Code 411 was the only sample totally misclassified—the sam-
ple was a 7-day cold sample, but was classified as a 6-day cold sample by
ANOVA and the standard curve. Codes 247 and 302 were not signifi-
cantly different from the 0-day cold or 4-day cold samples according to
Tukey-Kramer ANOVA analysis (See Figure 6). These two samples were
classified into either group. These results are not surprising, since the
sensory analysis and aerobic plate counts showed that the 0-day cold and
4-day cold samples were quite similar (see Table 4). Sensory analysis
classified the 0-day cold and 4-day cold samples as having attributes asso-
ciated with a mid quality sample. When spoilage was estimated using the
248               JOURNAL OF AQUATIC FOOD PRODUCT TECHNOLOGY

FIGURE 4. Ammonia standard curve. Both linear and exponential
equations are plotted on the same graph. The exponential equation better
fits the data as is reflected in the R2 value.

                                       Ammonia Standard                  y = 0.3969
            100
                                                                          R2 = 0.9175
             90
             80
             70
             60
                                                                       y = 11.118x –
             50
 ppm

                                                                           R2 = 0.8075
             40
             30
             20
             10
              0
                  0       1      2     3       4       5           6        7          8
            –10
                                              Day

FIGURE 5. Ammonia data analyzed by the Tukey-Kramer test for One-Way
ANOVA. All groups were found to be significantly different at the α = 0.05
level.

            100

             75
      ppm

             50

             25

              0
                      day 0   day 4   day 5    day 6       day 7       All Pairs
                                                                       Tukey-Kramer
                                                                       0.05
                                      day
Sarnoski et al.                                249

                     TABLE 3. Draeger coded unknown results

Code           Parts Per        Estimated Day          Actual Day        ANOVA
              Million (ppm)    (Standard Curve)                     Classification (Day)

411                71                6.4                   7                6
949                55                6.0                   5                5
247                 8                3.1                   0                0/4
455                85                6.6                   7                7
339                41                5.6                   5                5
302                10                3.4                   4                0/4

Parts per million (ppm) values were determined by Draeger-Tubes analysis and standard
curve (exponential) was used to compute the estimated day. JMP IN was used for the
ANOVA classification

FIGURE 6. Ammonia data analyzed by the Tukey-Kramer test for One-Way
ANOVA with coded unknowns. A means comparison analysis was
performed on this data and the statistical procedure identified into which
spoilage day each coded unknown should be classified.

       90
       80
       70
       60
       50
 ppm

       40
       30
       20
       10
        0
            247 302 339 411 455 949 day 0 day 4 day 5 day 6 day 7   All Pairs
                                                                    Tukey-Kramer
                                                                    0.05
                                     day

standard curve, code 247 was estimated to be a 3.1-day sample (closer to
0 day), and code 302 was estimated to be a 3.4-day sample (closer to a
4 day), thus corresponding closely with the actual spoilage increment.
Most importantly, both samples were classified correctly as “pass.” What
is important about this method is the sensitivity of the Draeger-Tube
250       JOURNAL OF AQUATIC FOOD PRODUCT TECHNOLOGY

           TABLE 4. Aerobic plate counts (APC) and sensory
          results of samples used for Draeger-Tubes analysis

          Sample                APC (CFU/g)           Sensory
                                                    Classification

          Reference (0 day)      1.65 • 105       Mid pass
          4 day cold              8.5 • 105       Mid Pass
          5 day cold              1.6 • 106       Low pass
          6 day cold              3.2 • 106       Low pass/low fail
          7 day cold                1 • 107       Mid fail/high fail

method compared to the electronic nose method. This method may be suf-
ficiently sensitive (at least for crabmeat) to separate crabmeat quality into
different categories. Thus, the coded samples 247 and 302 were consid-
ered to be correctly classified.
   The Draeger-Tubes rely on a simple chemical reaction. The tubes con-
tain a pH indicator. The reaction principle is that as NH3 (a base) is
exposed to the pH indicator, a color change occurs due to a rise in pH,
demonstrated by a change in color from yellow to blue inside the tube.
Perhaps this reaction can still be used, but it may need to be modified to
develop a dipstick test that industry could easily use. The Draeger method
is simple and more rapid compared with the chemical determination of
ammonia and total volatile bases. The technology may also be easily
transferable to industry because many seafood processors already use
Draeger-Tubes to check for ammonia refrigerant leaks.
Comparison of Draeger-Tube and Electronic Nose Methods
   It is important to note that the training samples and the samples used
for validation (coded unknown samples) were from 2 different batches.
The sample groups (0, 4, 7 day cold, etc.) for training and validation
were of the same or similar spoilage time, but they were from different
crabmeat batches. Duplicate samples were run during training and vali-
dation and yielded different results using the electronic nose (refer to
Figure 2). Draeger-Tube and sensory analysis were able to recognize
similar spoilage groups across different batches of crabmeat; however
the electronic nose often misclassified or even classified replicate sam-
ples into different spoilage groups. This suggests that the Draeger-Tube
procedure was more robust than the electronic nose under conditions of
use in this study.
Sarnoski et al.                                   251

                                 CONCLUSIONS

  The results for the coded validation samples indicate that if the Cyranose
320 electronic nose is used as a quality indicator for crabmeat, improve-
ment in its technology is needed. The results indicate that the Draeger-
Tubes show promise for determining spoilage levels for crabmeat and
could be used as a quality control procedure by industry. More research is
needed to determine if the method works across many batches, seasons,
environments, and species.

                                  REFERENCES

Arshak, K., Moore, E., Lyons, G. M., Harris, J., and Clifford, S. 2004. A review of gas
   sensors employed in electronic nose applications. Sensor Rev. 24: 183–195.
Balasubramanian, S., Panigrahi, S., Logue, C. M. Marchello, M., Doetkott, C., Gu, H.,
   Sherwood, J., and Nolan, L. 2004. Spoilage identification of beef using an electronic
   nose system. Trans. ASAE. 47: 1625–1633.
Bonzek, C., Fegley, L., Hoenig, J., Miller, T., O’Reilly, R., Orner, D., Sharov, A.,
   Terceiro, M., and Vaughan, D. 2005. 2005 Chesapeake Bay Blue Crab Advisory
   Report. Chesapeake Bay Stock Assessment Committee.
Chantarachoti, J., Oliveria, A. C. M., Himelbloom, B. H., Crapo, C. A., and McLachlan,
   D. G. 2006. Portable electronic nose for detection of spoiling Alaska pink salmon
   (Onchorhynchus gorbuscha). J. Food Sci. 71: S414–S421.
Cyrano Sciences. 2001. The Practical Guide to the Cyranose 320. Rev. C. Pasadena, CA:
   Cyrano Sciences, Inc.
Dräger-Rohrchen. 2001. Instructions for Use. 9th ed. Luebeck, Germany: Dräger.
Eskin, N. A., Henderson, H. M., and Townsend, R. J. 1971. Biochemistry of foods.
   New York: Academic Press, Inc.
FDA. 2001. Bacteriological Analytical Manual Online. 8th ed.: U.S. Food and Drug
   Administration.
Gardner, J., and Bartlett, P. 1999. Electronic Noses Principles and Applications.
   New York: Oxford University Press.
Gill, T. 1990. Objective analysis of seafood quality. Food Rev. Int. 6: 687–697.
Harper, W. J. 2001. The strengths and weaknesses of the electronic nose. Adv. Exp. Med.
   Biol. 488: 59–71.
Harsanyi, G. 2000. Polymer films in sensor applications a review of present uses and
   future possibilities. Sensor Rev. 20: 71–81.
Hines, E. L., Llobet, E., and Gardner, J. W. 1999. Electronic noses: a review of signal pro-
   cessing techniques. Circuits, Devices and Systems, IEE Proceedings 146: 297–310.
Hu, X., Quillin, R. C., Matanin, B. M., Cheng, B., Mallikarjunan, P., and Vaughan, D.
   2005. Development of non-destructive methods to evaluate oyster quality by electronic
   nose technology. ASAE Paper No. 056097. St. Joseph, MI: ASAE.
252         JOURNAL OF AQUATIC FOOD PRODUCT TECHNOLOGY

Jones, P. R. H., Ewen, R. J., and Ratcliffe, N. M. 1998. Simple methods for the extraction
   and identification of amine malodurs from spoiled foodstuffs. J. Food Compos. Anal.
   11: 274–279.
Marr, A. G. 2000. Growth kinetics, bacterial. Lederberg, J., ed. In: Encyclopedia of
   Microbiology. 2nd ed. San Diego: Academic Press. pp. 584–589.
Martin, R., Flick, G., Hebard, C., and Ward, D. 1982. Chemistry and Biochemistry of
   Marine Food Products. Westport, CT: AVI Publishing Co. pp. 137–368
Meilgaard, M., Civille, G. V., and Carr, B. T. 1999. In: Sensory Evaluation Techniques.
   3rd ed. Boca Raton, FL: CRC Press. pp. 168–172.
O’Connell, M., Valdora, G., Peltzer, G., and Martin Negri, R. 2001. A practical approach
   for fish freshness determinations using a portable electronic nose. Sens. Actuators B
   80: 149–154.
Pearce, T. C., Schiffman, S. S., Nagle, H. T., and Gardner, J. W. 2003. Handbook of
   Machine Olfaction. Weinheim: Wiley-VCH.
Steinbrecher, K. 1973. Collaborative study of the determination of ammonia as an index
   of decomposition in crabmeat. J. AOAC Int. 56: 598–599.
Van Deventer, D., and Mallikarjunan, P. 2002. Comparative performance analysis of three
   electronic nose systems using different sensor technologies in odor analysis of retained
   solvents on printed packaging. J. Food Sci. 67: 3170–3183.
Yerlikaya, P. and Gökoðlu, N. 2004. Quality changes of blue crab (Callinectes sapidus)
   meat during frozen storage. J. Food Qual. 27: 83–89.
Zhang, H., Balaban, M. Ö., and Principe, J. C. 2003. Improving pattern recognition of
   electronic nose data with time-delay neural networks. Sens. Actuators B 96: 385–389.
Zhang, H., Balaban, M. Ö., Principe, J. C., and Portier, K. 2005. Quantification of spice
   mixture compositions by electronic nose: Part 1. Experimental design and data analysis
   using neural networks. J. Food Sci. 70: E253–258.
You can also read